Spatial Temporal Adaptive Direct Mode Coding on Bidirectional Motion Compensation Prediction

摘要

Although bidirectional motion compensation reduces the coding distortion of a predictive image and improves prediction efficiency, the information indicating the spatial and temporal position of the predictive image is overhead and consumes more bits. Consequently, the coding efficiency is not always improved. The direct mode is widely used to improve coding efficiency with small overhead since it calculates the motion vectors of bidirectional prediction using the time continuity of motion. Such an improvement in prediction efficiency, however, is not achieved when the time continuity of motion does not exist. Therefore, the efficiency is highly dependent on the contents of a sequence. In this paper, we propose a spatial temporal adaptive direct mode, which is efficient even when there is no time continuity of motion. In this method, the time continuity of motion is predicted from the coding states of spatially or temporally adjacent blocks, and either a temporal direct mode or a spatial prediction mode is chosen as the coding mode. The spatial prediction mode calculates the motion vectors and the prediction direction from the motion vectors of the spatially adjacent blocks. This paper discusses the problem of the conventional direct mode, and proposes its solution. The results of computer simulations prove that this solution method can select the appropriate coding mode adaptively for the contents of a sequence, thus achieving high coding efficiency.